It started with a simple goal: I wanted an AI to learn my writing voice.
I'd been publishing on my WeChat account for eight years — scattered posts since arriving in the US in 2018. Late-night fragments, book notes, industry analysis, personal reflections. 170 pieces in total, written on and off. Some were crafted with care; others were just emotional vents at 2 a.m. I wanted the AI to absorb my habits of expression so that when it helped me polish a draft, it wouldn't turn me into someone else.
What it returned was far more than a style summary.
It gave me a fifteen-dimension analysis: thinking framework, value system, worldview, cognitive traits, emotional map, cultural identity, and more. I stared at it for a long time. Not because it was well written — but because some of it understood me better than I understood myself.
So I kept feeding it data. More weekly journals, more fragments, more reading notes. I wanted to see what would happen if I kept going.
Somewhere in the process, a thought struck me: if an AI can "understand" a person to this degree from text alone, does that mean we can use this to remember the people close to us?
I thought of my grandmother. If I had collected more material back then — the way she spoke, her catchphrases, her perspective on things — could I still "talk" with her after she was gone?
I'd had this idea since 2020. But only after being analyzed by AI myself did I truly grasp what it meant.
So I decided to go first. I'd open-source myself.
What It Saw
After feeding it 170 pieces, the AI did something I hadn't expected: it didn't summarize them one by one. It went across the timeline, searching for relationships between things written at different times.
Any single weekly journal is just a piece of scattered reflection. But put 170 of them together, and the AI extracted something humans aren't particularly good at: cross-temporal patterns.
Before showing the specific findings, here's the big picture it painted.
Across eight years of writing, it identified a clear line of intellectual evolution:
- 2018–2020: Idealism and anxiety coexisting. Just arrived in the US, questioning the meaning of life, searching for the optimal path. The grand ambition to "carry forward the lost teachings of past sages" existed alongside the inferiority of having attended "a bottom-tier 985 university." A spiritual void, and the beginning of self-rescue through writing and reading.
- 2021–2022: "Happiness must come from within" — finding a spiritual anchor. Discovering that emotions are a valuable resource. Learning not to judge others. Expanding from self-cultivation to understanding others.
- 2023: A period of discipline and rhythm. 52 uninterrupted weekly journals. 57 books read. "Writing is the beginning of perceiving the world again."
- 2024: Loosening and awakening. Growing doubts about the meaning of work. The J-type personality began to soften. Meeting YR, moving from "island" to "relationship." The journal title changed from "Summary" to "Ramblings."
- 2025: "Pessimists are right; optimists succeed." Started exploring AI content creation in spare time. Proposed. "Doing the right thing matters more than doing things right."
The overview was interesting. But what truly startled me was what the AI saw when it zoomed in — the thin lines I hadn't noticed myself.
Finding One: The Decision Chain Behind My Proposal
The AI told me: You're an extremely rational person, but you made a "very sudden" emotional decision — proposing.
My first reaction: Yeah, it was sudden.
Then the AI said: No. Looking back at your writing, it wasn't sudden at all.
It strung together a complete chain from weekly journals scattered across half a year:
October 2024. I attended a friend's wedding. That day I wrote: "Love is a decision, not a feeling." The groom's vow was "From now on, you are my Top priority." I wrote: "This man is my role model."
Later that same month, after a grueling 29-kilometer hike, I wrote: "Maybe in the long journey of life, very little is actually within our control. Most of it isn't." — For someone used to controlling everything, this was a significant loosening.
April 2025. In a reading note, I quoted a line: "Only after deep conviction does happiness come. Only after deep conviction is there the courage to take the next step."
May 2025. I proposed. In my journal: "The fundamental reason is that during the ten days we were apart, we both realized the other had already merged into our lives. There was no reason not to be together. So that's how it was decided."
When I read this analysis, I was genuinely stunned. I hadn't realized there was such a timeline — from "love is a decision" to "accepting loss of control" to "happiness comes after conviction" to "so that's how it was decided."
I thought proposing was a sudden decision. The AI told me I had spent more than half a year rationally persuading myself to accept "emotional conviction."
Finding Two: The Slow Loosening of a J-Type
The second surprise was about my personality slowly loosening.
The AI said: Your writing records the "unclenching" of a classic J-type personality.
Early 2024. I started tracking my time usage in extreme detail — what I was doing every hour. My mother is a teacher; every summer and winter break since childhood, she set goals for me. I inherited that habit. But in the same journal, I wrote one line: "And it seems both of us have recently realized that maybe life doesn't have to be this packed all the time."
Mid-2024. I started naming my weekly journals "Ramblings." Before that, they were all "Week X Summary" — the naming itself was becoming looser.
Late 2024. I wrote: "An unproductive weekend, with some guilt." — But right after, I began questioning the guilt itself.
Early 2025. YR and I were running late at Disney. I thought of Liu Liangcheng's line: "There's no need to rush through life," and added in my journal: "Especially when you're out having fun."
Mid-2025. I wrote: "In middle and high school, I hoped to at least make a splash in the long river of history. Now I'm increasingly clear that I'm just an ordinary person."
If the AI hadn't strung this line together, I probably wouldn't have noticed: it took me a year and a half to move from "every minute must be meaningful" to "there's no need to rush through life."
The change was so slow that even the person living it couldn't feel it. But in 170 pieces of writing, the AI saw it.
Gradual change is the hardest kind for the person involved to notice — but AI can.
Finding Three: How a Reading Note Became a Decision Six Months Later
In March 2025, I wrote a reading note about Shaonan, author of the newsletter Product Contemplations.
"Shaonan is the kind of person I want to become — constantly learning, distilling knowledge, sharing it with people who share the same interests."
On its own, it's just a reader's reflection. But the AI placed it alongside writing from the months before and after, and saw a bigger picture.
Looking back:
November 2024. A project finally wrapped up. But I couldn't feel any joy. In my journal: "Everyone else didn't seem to be cheering either... We're just Sisyphus, endlessly pushing the stone — I mean, the project." Around the same time, I finished The Remains of the Day and wrote: "The butler missed his father's final moments and his life partner for the sake of 'great events'... Working hard is fine, but sacrificing too much for work is a losing trade."
March 2025. I wrote an "epitaph" for the team I'd been on for four years — "No one accompanied it through its entire lifecycle." Then: "Work is one fulcrum in life, but not the only one."
April 2025. More specific: "Shaonan sits in his study for 8–10 hours on Sundays, reading, distilling knowledge, writing 5,000 to 10,000 words... That's exactly what I want to do."
Looking forward: a few months later, I started building an AI content team in my spare time — doing exactly what Shaonan does: learning, distilling knowledge, sharing.
This isn't a heroic story of quitting to start a company. I didn't leave my job. I just realized the current work no longer attracted me, and I wanted to find a passion — something that could hook life and work together again. And that reading note — "Shaonan is the kind of person I want to become" — in hindsight, was an unconscious blueprint I'd drawn for myself.
The AI saw something in that reading note that I hadn't yet seen.
Finding Four: The Seed of Digital Resurrection
The last finding has to do with this essay itself.
In the writing from 2024–2025, the AI found a thematic thread — an emotion that recurred across different months and different topics: cherishing the connections between people.
July 2024. YR and I were walking under a half-dead tree. She said: "I wonder if death is another kind of rebirth." I said: "Maybe this life is a dream." She said: "Then when you wake up, you must remember to come find me again." We made a pinky promise with tears in our eyes.
Around the same time, I set up a NAS system at home to manage photo backups, using the 3-2-1 backup rule to make sure no photo would ever be lost.
Late July 2024. We watched Robot Dreams. YR cried and said: "I don't want to be just a passerby in your life." I wrote in my journal: "No matter how many times I write this, tears well up."
January 2025. I had three different AIs polish the same passage and compared which one better preserved my way of expressing myself — which was itself the earliest experiment in "making AI understand human expression."
The AI strung these together and interpreted them as "memory anxiety" — a fear of loss.
But honestly, that reading isn't quite right.
The idea of digital resurrection came to me in 2020. Its foundation isn't anxiety; it's cherishing — I cherish the connections with the people around me. If I could collect more material about my grandmother, I could still miss her after she's gone. The pinky promise about the next life isn't fear; it's love. The NAS isn't anxiety; it's caring.
The AI saw "what kept recurring," but it got "why it kept recurring" wrong.
And that raises a more interesting question: how deeply can AI really understand a person?
What It Couldn't See
When I told the AI that its interpretation of my motivation was off, it did something interesting: it didn't simply say "I was wrong." It immediately absorbed my correction into its analytical framework and explained it as "AI's tendency toward over-interpretation."
There was another layer of interest in that — AI is always looking for patterns, even in its own mistakes.
That made me think seriously about the boundaries.
What did it get right? It's good at finding patterns — recurring themes across time, consistency behind scattered signals, gradual changes the person themselves can't perceive. 170 pieces is too much for a human. No one goes back to compare a journal from three years ago with yesterday's. But AI does.
What did it get wrong? Motivation. The AI elevated "exploration" to "life pivot," read "cherishing" as "anxiety." It naturally tends to string scattered signals into an arc with direction — the threading is accurate, but the "meaning" assigned to the arc may not be.
Beyond that, there are three more fundamental things AI can't see.
First, it can't see what was never written down.
In 170 pieces, I wrote a lot about work, reading, YR. But I almost never wrote about: late-night deep conversations with friends, the details of a technical decision at work that made me proud, the reconciliation after a fight with YR.
The "me" the AI analyzed is a me filtered through writing. The real me is more complex, more contradictory, messier than what appears in text.
Second, it can't distinguish "what I really thought" from "what I wrote down."
When people write journals, they unconsciously beautify, organize, assign meaning. When I wrote "love is a decision," did I really think that in that moment? Or did I add a layer of meaning in retrospect? AI can't tell.
Third, it doesn't know what "silence" means.
In 2024, there were weeks when I didn't write. Was I too busy? Too happy to need writing? Or too sad to write? Silence may be the most informative signal — but AI can't read silence.
Replicable, and Not
Back to the original question: if I hand 170 pieces of writing to an AI, can it "replicate" me?
Honest answer: yes and no.
The yes part: The AI did capture my thinking patterns — how I approach problems, what I care about, what I fear, my decision logic. If you ask "this AI version of me" a question, its answer might be 70–80% like the real me.
The no part: It doesn't have the moment under the tree in the cemetery, making a pinky promise with tears. It doesn't have the bodily sensation of choking up at 2 a.m. while writing about my grandmother. It has my thinking framework, but not my body heat.
Last year, YR celebrated my birthday. She didn't book a fancy restaurant — she booked a small wooden cabin, made longevity noodles, and prepared a Little Prince cake. I wrote in my journal: "I'm happy. I met someone who understands me and is willing to love me in the way I hoped for."
That's the ultimate form of "being understood" — it doesn't take fifteen analytical dimensions. It just takes one person who knows you want a cabin, not a Michelin star.
AI can analyze you, but it can't understand you. "Analyzing" and "understanding" are two different things.
So I Keep Writing
I open-sourced myself. 170 pieces, eight years of inner monologue, a fifteen-dimension analysis report.
The AI told me things I hadn't noticed — my proposal wasn't sudden, my exploration of new things wasn't a whim, my cherishing of human connection is deeper and more persistent than I thought.
It also got some things wrong — reading cherishing as anxiety, reading exploration as a pivot. And when I corrected it, it tried to absorb the correction into its analysis too.
But the AI also reminded me of something: what it can preserve is the trace of thought, not the thought itself. Just as a NAS can back up photos but not the feeling when they were taken.
Maybe one day, digital resurrection technology really will be able to replicate a person's way of thinking — letting the deceased "come back" in some form. I was thinking about this in 2020, and I'm still thinking about it now.
But until that day comes, what I can do is keep writing.
Not for the AI, but for my future self — I hope that when he looks back, he can recognize from these words the person who colored with crayons in a pizza shop, made a tearful pinky promise in a cemetery, and wrote weekly journals at 2 a.m.
170 pieces can't replicate a person. But they can prove this person existed, lived seriously, and thought.